Add Pythonic SageMaker training and processing

[rohitrsh]

Stacked PR 2/2 — do not merge out of order.

This branch currently includes the commit from #3456. After #3456 merges, this branch will be rebased onto master so the final review contains only the Pythonic SageMaker change.

Summary

• Adds Pythonic SageMaker Training and Processing tasks that execute ordinary Flyte @task functions inside SageMaker-managed containers.
• Introduces shared entrypoint, environment, job naming, output, error, retry, and cancellation handling for Pythonic SageMaker jobs.
• Requires an explicit container image and region, limits Pythonic jobs to one instance, validates SageMaker entrypoint limits, and rejects incompatible Processing network isolation.
• Documents image requirements, IAM behavior, model artifacts, inputs/outputs, and the distinction between declarative and Pythonic modes.
• Treats RETRYABLE_FAILED as terminal for local connector execution and adds regression coverage.

Stack order:

1. Add managed SageMaker job connectors #3456 — Managed SageMaker job connectors
2. Pythonic SageMaker Training and Processing (this PR)

Test plan

• Focused Pythonic and managed Training/Processing connector suite — 74 passed
• Core connector utility suite — 15 passed
• SageMaker plugin lint and formatting checks
• Plugin packaging, serialization, and import smoke tests
• DCO, confidentiality, attribution, and internal-reference audits

Merge only after #3456. Before merge, rebase this branch onto upstream master, rerun all local and hosted verification gates, and mark the PR ready for review.